Joint analysis of galaxy clustering and weak lensing via simulation-based inference

物理宇宙学被一个成功的连接高能物理和观测天文学的标准模型所主导，在这个模型中，宇宙从大爆炸演化到现在的状态。然而，我们目前还不知道中微子粒子加速膨胀和质量的原因。对膨胀加速度的精确测量将检验广义相对论的有效性，并揭示暗能量的物理本质，暗能量是一种在最大范围内影响宇宙的未知能量形式。测量中微子的未知质量，长期以来一直被认为是零，是阐明它们的本质和探索超越粒子物理标准模型的新物理学的重要一步。这个研究计划解决了这两个令人着迷的问题。回答这样的物理问题需要从大量的天文数据集中以足够的精度提取信息。这是一个复杂的问题，因为当人们改变模型而不是标准模型时，可观测性的变化是非常微妙的。需要先进的技术来梳理出物理学。然而，现有的方法依赖于各种简化和假设，在某种程度上是不成立的。作为回应，我的技术代表了对如何分析数据的彻底反思，将物理定律和仪器的工作原理纳入数据分析中使用的计算机模型。这在过去几年里才有可能做到，这要归功于我引入的新方法，包括大幅减少所需模型评估数量的统计算法，以及使这些评估大量并行的计算物理技术。这些发展形成了我将要遵循的原则性方法的基础，即所谓的基于模拟的推理。使用物理计算机模型的基于模拟的推理将使我第一次能够独一无二地利用宇宙大尺度结构的完整观测地图，从而超越使用所谓的相关函数的标准估计器所获得的结果。因此，使我能够实现科学目标的关键创新是同时分析在大规模结构调查中可观察到的两个主要影响：星系群聚和弱引力透镜。这一进展在欧空局的欧几里德卫星大规模观测项目中尤为及时和重要，因为它将为其核心方案的两个方面提供独特的见解。使用物理计算机模型来联合分析星系团和弱引力透镜将减少统计和系统的不确定性，提高宇宙学结果的精度，并相对于基于关联函数的标准方法提高其精度。我的建议包括开发、验证和应用新的用于星系调查数据分析的物理计算机数据模型，以满足这两个科学目标。这项研究构成了从大规模天文数据中提取物理信息的概念上的全新方法。这项拟议的工作将首先对合成的欧几里德数据进行联合聚类透镜分析，然后通过基于模拟的推理对其进行分析。这些分析将产生参考质量的宇宙学信息，由下列交付成果提供：星系团和弱引力透镜对膨胀加速的联合约束，以及中微子质量。该研究将在帝国推理和宇宙学中心主办。

Physical cosmology is dominated by a successful standard model connecting high-energy physics to observational astronomy, in which the Universe evolves from the "Big Bang" to its present state. However, we currently do not know the cause of the accelerated expansion and the masses of neutrino particles. Precise measurements of the acceleration of expansion will test the validity of general relativity and shed light on the physical nature of dark energy, an unknown form of energy that affects the Universe on the largest scales. Measuring the unknown masses of neutrinos, long thought to be zero, is an important step towards elucidating their essential nature and exploring new physics beyond the standard model of particle physics. This research programme tackles these two enthralling problems.Answering such physical questions requires extracting information from large astronomical data sets with sufficient accuracy. This is a complex problem, since changes to observables when one changes the model away from the standard model are extremely subtle. Advanced techniques are required to tease out the physics. However, existing methods rely on various simplifications and assumptions that at some level do not hold. For the next generation of galaxy surveys, it will be vital to improve upon these.As a response, my techniques represent a radical rethinking of how to analyse data, incorporating of the laws of physics and the working of instruments into computer models used within data analysis. This has only been possible to do at all in the last few years thanks to novel methods that I introduced, including statistical algorithms that drastically reduce the number of model evaluations required, and computational physics techniques that make these evaluations massively parallel. These developments form the basis of the principled methodology that I will follow, known as 'simulation-based inference'.Simulation-based inference using physical computer models will uniquely enable me, for the first time, to exploit complete observed maps of the large-scale structure of the Universe and, thus, to go beyond the results obtained with standard estimators of so-called 'correlation functions'. Consequently, the key innovation that will allow me to address the science goals is a simultaneous analysis of the two main effects observable in large-scale structure surveys: galaxy clustering and weak gravitational lensing. This development is particularly timely and significant in the context of the large-scale observational project of ESA's Euclid satellite, as it will provide unique insights into both aspects of its core programme. The use of physical computer models to jointly analyse galaxy clustering and weak gravitational lensing will reduce statistical and systematic uncertainties, increase the precision of cosmological results, and improve their accuracy with respect to standard methods based on correlation functions.My proposal consists of developing, validating, and applying new physical computer data models for galaxy survey data analysis, in order to address the two scientific objectives. This research constitutes a conceptually entirely new approach to extracting physical information from large-scale astronomical data. The proposed work will culminate with the first joint clustering-lensing analyses of synthetic, then real Euclid data via simulation-based inference. These analyses will produce cosmological information of reference quality, provided by the following deliverables: joint constraints from galaxy clustering and weak gravitational lensing on the acceleration of the expansion, and on neutrino masses.The fellowship will be hosted at the Imperial Centre for Inference and Cosmology.